
from skimage.metrics import peak_signal_noise_ratio as psnr
import matplotlib.pyplot as plt
import cv2
import os
import numpy as np
import pandas as pd
import kagglehub
from ultralytics import YOLO

/home/mperpinav/.local/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html
  from .autonotebook import tqdm as notebook_tqdm


# Nos traemos la BD CBIS-DDSM

path_cbis_ddsm = kagglehub.dataset_download("awsaf49/cbis-ddsm-breast-cancer-image-dataset")

print("Path to dataset files:", path_cbis_ddsm)

Warning: Looks like you're using an outdated `kagglehub` version (installed: 0.3.11), please consider upgrading to the latest version (0.3.12).
Path to dataset files: /home/mperpinav/.cache/kagglehub/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset/versions/1


# Cargamos los csv
dicom_data = pd.read_csv(os.path.join(path_cbis_ddsm, 'csv/dicom_info.csv'))
image_dir = os.path.join(path_cbis_ddsm, 'jpeg')
dicom_data.head()


dicom_data.SeriesDescription.value_counts()

SeriesDescription
cropped images           3567
ROI mask images          3247
full mammogram images    2857
Name: count, dtype: int64


cropped_images = dicom_data[dicom_data.SeriesDescription=="cropped images"].image_path
cropped_images.head()

0     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.129308...
3     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.381187...
6     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.153339...
7     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.178994...
10    CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.411833...
Name: image_path, dtype: object


cropped_images = cropped_images.apply(lambda x: x.replace('CBIS-DDSM/jpeg', image_dir))
cropped_images.head()

0     /home/mperpinav/.cache/kagglehub/datasets/awsa...
3     /home/mperpinav/.cache/kagglehub/datasets/awsa...
6     /home/mperpinav/.cache/kagglehub/datasets/awsa...
7     /home/mperpinav/.cache/kagglehub/datasets/awsa...
10    /home/mperpinav/.cache/kagglehub/datasets/awsa...
Name: image_path, dtype: object


full_mammogram_images = dicom_data[dicom_data.SeriesDescription == 'full mammogram images'].image_path
full_mammogram_images.head()

1     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.248386...
2     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.267213...
11    CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.210396...
12    CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.749566...
15    CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.987658...
Name: image_path, dtype: object


full_mammogram_images = full_mammogram_images.apply(lambda x: x.replace('CBIS-DDSM/jpeg', image_dir))
full_mammogram_images.head()

1     /home/mperpinav/.cache/kagglehub/datasets/awsa...
2     /home/mperpinav/.cache/kagglehub/datasets/awsa...
11    /home/mperpinav/.cache/kagglehub/datasets/awsa...
12    /home/mperpinav/.cache/kagglehub/datasets/awsa...
15    /home/mperpinav/.cache/kagglehub/datasets/awsa...
Name: image_path, dtype: object


ROI_mask_images = dicom_data[dicom_data.SeriesDescription == 'ROI mask images'].image_path
ROI_mask_images.head()

5     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.153339...
8     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.178994...
9     CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.411833...
14    CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.236373...
20    CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.357008...
Name: image_path, dtype: object


ROI_mask_images = ROI_mask_images.apply(lambda x: x.replace('CBIS-DDSM/jpeg', image_dir))
ROI_mask_images.head()

5     /home/mperpinav/.cache/kagglehub/datasets/awsa...
8     /home/mperpinav/.cache/kagglehub/datasets/awsa...
9     /home/mperpinav/.cache/kagglehub/datasets/awsa...
14    /home/mperpinav/.cache/kagglehub/datasets/awsa...
20    /home/mperpinav/.cache/kagglehub/datasets/awsa...
Name: image_path, dtype: object


# Creamos una función para corregir el path de las imagenes
def replace_path(sample, old_path, new_path):
    return sample.replace(old_path,new_path,regex=True)


from PIL import Image
def plot_smaples(sample, row=15, col=15):
    plt.figure(figsize=(row, col))
    for i, file in enumerate(sample[0:5]):
        cropped_images_show = Image.open(file)
        gray_img= cropped_images_show.convert("L")
        plt.subplot(1,5,i+1)
        plt.imshow(gray_img, cmap='gray')
        plt.axis('off')
    plt.show()


plot_smaples(cropped_images, 15,15)


plot_smaples(full_mammogram_images,15,15)


plot_smaples(ROI_mask_images)


def get_image_file_name(data, new_dict):
    """
    /home
    /mperpinav
    /.cache
    /kagglehub
    /datasets
    /awsaf49
    /cbis-ddsm-breast-cancer-image-dataset
    /versions
    /1
    /jpeg
    /1.3.6.1.4.1.9590.100.1.2.153339052913121382622526066491844156138/2-270.jpg
    
    return path at index [12] after split depends on split('\')
    """
    for dicom in data:
        key = dicom.split('/')[11]
        new_dict[key] = dicom
    print(f"the length of dataset ==> {len(new_dict.keys())}")


cropped_images_dict = dict()
full_mammo_dict = dict()
roi_img_dict = dict()

get_image_file_name(cropped_images, cropped_images_dict)
get_image_file_name(full_mammogram_images, full_mammo_dict)
get_image_file_name(ROI_mask_images, roi_img_dict)

the length of dataset ==> 3567
the length of dataset ==> 2857
the length of dataset ==> 3247


next(iter((cropped_images.items())))

(0,
 '/home/mperpinav/.cache/kagglehub/datasets/awsaf49/cbis-ddsm-breast-cancer-image-dataset/versions/1/jpeg/1.3.6.1.4.1.9590.100.1.2.129308726812851964007517874181459556304/1-172.jpg')


def fix_image_path(data):
    """Correct dicom paths to correct image paths."""
    for indx, image in enumerate(data.values):
#         print(f"Image Path: {image[11]}")

        img_name = image[11].split('/')[2]
#         print(f"Looking for key: {img_name}")  # Debugging step

        if img_name in full_mammo_dict:
            data.iloc[indx, 11] = full_mammo_dict[img_name]
        else:
            data.iloc[indx, 11] = None
#             print(f"KeyError: '{img_name}' not found in full_mammo_dict")  # Debugging step
        
        img_name = image[12].split('/')[2]
        if img_name in cropped_images_dict:
            data.iloc[indx, 12] = cropped_images_dict[img_name]
        else:
            data.iloc[indx, 11] = None
            # print(f"KeyError: '{img_name}' not found in cropped_images_dict")  # Debugging step

        img_name = image[13].split('/')[2]
        if img_name in roi_img_dict:
            data.iloc[indx, 13] = roi_img_dict[img_name]

        else:
            data.iloc[indx, 13] = None


# Cargamos los csv
mass_train = pd.read_csv(os.path.join(path_cbis_ddsm, 'csv/mass_case_description_train_set.csv'))
mass_test = pd.read_csv(os.path.join(path_cbis_ddsm, 'csv/mass_case_description_test_set.csv'))

mass_train.head()


mass_train.iloc[:, 11].head()

0    Mass-Training_P_00001_LEFT_CC/1.3.6.1.4.1.9590...
1    Mass-Training_P_00001_LEFT_MLO/1.3.6.1.4.1.959...
2    Mass-Training_P_00004_LEFT_CC/1.3.6.1.4.1.9590...
3    Mass-Training_P_00004_LEFT_MLO/1.3.6.1.4.1.959...
4    Mass-Training_P_00004_RIGHT_MLO/1.3.6.1.4.1.95...
Name: image file path, dtype: object


mass_test.iloc[:, 11].head()

0    Mass-Test_P_00016_LEFT_CC/1.3.6.1.4.1.9590.100...
1    Mass-Test_P_00016_LEFT_MLO/1.3.6.1.4.1.9590.10...
2    Mass-Test_P_00017_LEFT_CC/1.3.6.1.4.1.9590.100...
3    Mass-Test_P_00017_LEFT_MLO/1.3.6.1.4.1.9590.10...
4    Mass-Test_P_00032_RIGHT_CC/1.3.6.1.4.1.9590.10...
Name: image file path, dtype: object


fix_image_path(mass_train)


mass_train.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1318 entries, 0 to 1317
Data columns (total 14 columns):
 #   Column                   Non-Null Count  Dtype 
---  ------                   --------------  ----- 
 0   patient_id               1318 non-null   object
 1   breast_density           1318 non-null   int64 
 2   left or right breast     1318 non-null   object
 3   image view               1318 non-null   object
 4   abnormality id           1318 non-null   int64 
 5   abnormality type         1318 non-null   object
 6   mass shape               1314 non-null   object
 7   mass margins             1275 non-null   object
 8   assessment               1318 non-null   int64 
 9   pathology                1318 non-null   object
 10  subtlety                 1318 non-null   int64 
 11  image file path          1318 non-null   object
 12  cropped image file path  1318 non-null   object
 13  ROI mask file path       1318 non-null   object
dtypes: int64(4), object(10)
memory usage: 144.3+ KB


mass_train = mass_train.rename(columns={'left or right breast': 'left_or_right_breast',
                                        'image view': 'image_view',
                                        'abnormality id': 'abnormality_id',
                                        'abnormality type': 'abnormality_type',
                                        'mass shape': 'mass_shape',
                                        'mass margins': 'mass_margins',
                                        'image file path': 'image_file_path',
                                        'cropped image file path': 'cropped_image_file_path',
                                        'ROI mask file path': 'ROI_mask_file_path'})
mass_train.head(5)


fix_image_path(mass_test)


mass_test = mass_test.rename(columns={'left or right breast': 'left_or_right_breast',
                                        'image view': 'image_view',
                                        'abnormality id': 'abnormality_id',
                                        'abnormality type': 'abnormality_type',
                                        'mass shape': 'mass_shape',
                                        'mass margins': 'mass_margins',
                                        'image file path': 'image_file_path',
                                        'cropped image file path': 'cropped_image_file_path',
                                        'ROI mask file path': 'ROI_mask_file_path'})
mass_test.head(5)


print(f'Shape of mass_train: {mass_train.shape}')
print(f'Shape of mass_test: {mass_test.shape}')

Shape of mass_train: (1318, 14)
Shape of mass_test: (378, 14)


def display_images(dataset, column, number):
    """Displays images in dataset, handling missing files and converting formats."""
    
    # create figure and axes
    fig, axes = plt.subplots(1, number, figsize=(15, 5))
    
    # Loop through rows and display images
    for index, (i, row) in enumerate(dataset.head(number).iterrows()):
        image_path = row[column]
        
       # Check if image_path is valid (not None) and exists
        if image_path is None or not os.path.exists(image_path):
            # print(f"File not found or invalid path: {image_path}")
            continue
        
        image = cv2.imread(image_path)
        
        # Handle case when image can't be read
        if image is None:
            # print(f"Error reading image: {image_path}")
            continue
        
        # Convert BGR to RGB if needed (for correct color display)
        if len(image.shape) == 3 and image.shape[2] == 3:
            image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
        
        ax = axes[index]
        ax.imshow(image, cmap='gray' if len(image.shape) == 2 else None)
        ax.set_title(f"{row['pathology']}")
        ax.axis('off')
        print(np.array(image).shape)
    
    plt.tight_layout()
    plt.show()


# Se muestran algunas imagenes de ejemplo
print('Full Mammograms:\n')
display_images(mass_train, 'image_file_path', 5)
print('Cropped Mammograms:\n')
display_images(mass_train, 'cropped_image_file_path', 5)
print('ROI_mask:\n')
display_images(mass_train, 'ROI_mask_file_path', 5)

Full Mammograms:

(4808, 3024, 3)
(4800, 2656, 3)
(5491, 2986, 3)
(5491, 3046, 3)
(5491, 2911, 3)

Cropped Mammograms:

(515, 431, 3)
(292, 256, 3)
(466, 423, 3)
(427, 422, 3)
(399, 465, 3)

ROI_mask:

(4808, 3024, 3)
(4800, 2656, 3)
(5491, 2986, 3)
(5491, 3046, 3)
(5491, 2911, 3)


# Ahora del conjunto test
print('Full Mammograms:\n')
display_images(mass_test, 'image_file_path', 5)
print('Cropped Mammograms:\n')
display_images(mass_test, 'cropped_image_file_path', 5)
print('ROI_mask:\n')
display_images(mass_test, 'ROI_mask_file_path', 5)

Full Mammograms:

(4006, 1846, 3)
(5491, 2011, 3)
(5904, 3200, 3)
(5952, 3352, 3)
(5116, 2641, 3)

Cropped Mammograms:

(384, 385, 3)
(369, 328, 3)
(214, 195, 3)
(220, 225, 3)
(405, 376, 3)

ROI_mask:

(4006, 1846, 3)
(5491, 2011, 3)
(5904, 3200, 3)
(5952, 3352, 3)
(5116, 2641, 3)


# Para no hacer el preprocesamiento dos veces, unimos mass_train y mass_test
mass_data = pd.concat([mass_train, mass_test], axis=0)


# Mostramos como ha quedado la distribución 
fig,ax = plt.subplots()
fig.suptitle("Distribución clases conjunto mass")
num_mass = mass_data['pathology'].value_counts()
ax.bar( num_mass.index, num_mass, color = "#027a6d")
plt.show()


print("\nConteo mass:")
print(num_mass)

Conteo mass:
pathology
MALIGNANT                  784
BENIGN                     771
BENIGN_WITHOUT_CALLBACK    141
Name: count, dtype: int64


# Se juntan las clases BENIGN y BENIGN_WITHOUT_CALLBACK para tener una distribución binaria.
# Unificacion de las clases BENIGN y BENIGN_WITHOUT_CALLBACK como BENIGN
mass_data['pathology'] = mass_data['pathology'].replace('BENIGN_WITHOUT_CALLBACK', 'BENIGN')


print("\nConteo mass:")
num_mass = mass_data['pathology'].value_counts()
print(num_mass)

Conteo mass:
pathology
BENIGN       912
MALIGNANT    784
Name: count, dtype: int64


# Mostramos como ha quedado la distribución 
fig,ax = plt.subplots()
fig.suptitle("Distribución clases conjunto mass")
ax.bar( num_mass.index, num_mass, color = "#027a6d")
plt.show()


# Guardamos los tipos de anomalia que tenemos
anomalia_tipes = []
for i in mass_data['pathology'].unique():
  anomalia_tipes.append(i)
print(anomalia_tipes)

['MALIGNANT', 'BENIGN']


mass_data.columns

Index(['patient_id', 'breast_density', 'left_or_right_breast', 'image_view',
       'abnormality_id', 'abnormality_type', 'mass_shape', 'mass_margins',
       'assessment', 'pathology', 'subtlety', 'image_file_path',
       'cropped_image_file_path', 'ROI_mask_file_path'],
      dtype='object')


count = 0
for i in range(len(mass_data)):
    img = Image.open(mass_data['image_file_path'].iloc[i])

    # Verificamos el modo de color
    if img.mode != 'L':  # 'L' es para escala de grises
        count += 1

print("Tenemos {} imágenes que no están en escala de grises".format(count))

Tenemos 0 imágenes que no están en escala de grises


import shutil

# Redimensionamos el tamaño de las imagenes del conjunto de entrenamiento
img_height = 640
img_width = 640

shutil.rmtree('Resized_mass_data', ignore_errors=True)
os.makedirs('Resized_mass_data') 
resize_mass_data = []
for i, path in enumerate(mass_data['image_file_path']):
    # Cargamos la imagen en escala de grises
    img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)

    resized_img = cv2.resize(img,(img_height,img_width))
    # Guardamos las imagenes redimensionadas
    output_path = f'Resized_mass_data/resized_{i}.jpg'
    cv2.imwrite(output_path, resized_img)
    resize_mass_data.append(output_path)

# Guardamos en el dataset para tener el etiquetado
mass_data['resize_path'] = resize_mass_data


# Lo aplicamos a todo el conjunto de entrenamiento 
psnr_values = []
# Guardamos las imagenes filtradas en una nueva ruta
mass_red_ruido = []
shutil.rmtree('Red_Ruido_mass', ignore_errors=True)
os.makedirs('Red_Ruido_mass') 
for i, path in enumerate(mass_data['resize_path']):
    # Cargamos la imagen en escala de grises
    img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)

    # Aplicamos reducción de ruido
    img_filtered = cv2.fastNlMeansDenoising(img,3,3,7,21) # Dejamos los valores recomendados por CV

    # Calcular el PSNR entre la imagen original y la imagen filtrada
    current_psnr = psnr(img, img_filtered)
    # Guardar el PSNR para esta imagen 
    psnr_values.append(current_psnr)

    # Guardamos las imagenes preprocesadas
    output_path = f'Red_Ruido_mass/red_ruido_{i}.jpg'
    cv2.imwrite(output_path, img_filtered)
    mass_red_ruido.append(output_path)

mass_data['red_ruido_path'] = mass_red_ruido  
print("El PSNR medio para el conjunto mass es: {}".format(np.mean(psnr_values)))

El PSNR medio para el conjunto mass es: 44.574180066128434


print('Reducción de ruido:\n')
display_images(mass_data, 'red_ruido_path', 5)

Reducción de ruido:

(640, 640, 3)
(640, 640, 3)
(640, 640, 3)
(640, 640, 3)
(640, 640, 3)


shutil.rmtree('CLAHE_mass', ignore_errors=True)
os.makedirs('CLAHE_mass') 

# Guardamos las nuevas rutas
clahe_mass = []
# Recorremos todo el conjunto de entrenamiento
for i, path in enumerate(mass_data['red_ruido_path']):
    # Cargamos la imagen en escala de grises
    img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)

    # Aplicamos el preprocesamiento con CLAHE
    clahe = cv2.createCLAHE()
    cl1 = clahe.apply(img)

    # Guardamos las imagenes preprocesadas
    output_path = f'CLAHE_mass/clahe_{i}.jpg'
    cv2.imwrite(output_path, cl1)
    clahe_mass.append(output_path)

# Guardamos en el dataset para tener el etiquetado
mass_data['clahe_path'] = clahe_mass


print('Mejora de contraste con CLAHE:\n')
display_images(mass_data, 'clahe_path', 5)
print('Imagen con reducción de ruido:\n')
display_images(mass_data, 'red_ruido_path',5)

Mejora de contraste con CLAHE:

(640, 640, 3)
(640, 640, 3)
(640, 640, 3)
(640, 640, 3)
(640, 640, 3)

Imagen con reducción de ruido:

(640, 640, 3)
(640, 640, 3)
(640, 640, 3)
(640, 640, 3)
(640, 640, 3)


from sklearn.model_selection import train_test_split

# Hacemos los conjuntos de train y test
X_mass = mass_data.drop(columns=['pathology'])
y_mass = mass_data['pathology']

X_train_mass, X_val_mass, y_train_mass, y_val_mass = train_test_split(
    X_mass, y_mass,
    test_size=0.2,
    random_state=42,
    stratify=y_mass
)

# Nos guardamos los datasets
mass_train = pd.concat([X_train_mass, y_train_mass], axis=1)
mass_val = pd.concat([X_val_mass, y_val_mass], axis=1)


# Comprobamos que la distribución entre conjunto train y test es correcta
fig, ax = plt.subplots()

fig.suptitle("Distibución clases conjunto mass train vs val")

num_mass_train = mass_train['pathology'].value_counts()
num_mass_val = mass_val['pathology'].value_counts() 

ax.bar(num_mass_train.index,num_mass_train, color = "#027a6d",  label = 'train' )
ax.bar(num_mass_val.index, num_mass_val,  color = "#44c27f", label = 'val')
plt.legend()
plt.show()


print("Distribución del conjunto train mass:")
for clase, num in (100 * mass_train['pathology'].value_counts() / (mass_train['pathology'].value_counts() + mass_val['pathology'].value_counts())).items():
    print(f"{clase}: {num:.2f}%")
print("Distribución del conjunto val mass:")
for clase, num in (100 * mass_val['pathology'].value_counts() / (mass_train['pathology'].value_counts() + mass_val['pathology'].value_counts())).items():
    print(f"{clase}: {num:.2f}%")

Distribución del conjunto train mass:
BENIGN: 79.93%
MALIGNANT: 79.97%
Distribución del conjunto val mass:
BENIGN: 20.07%
MALIGNANT: 20.03%


# yolo espera que las etiquetas de clase sean numericas
label_map = {'BENIGN': 0, 'MALIGNANT': 1}
mass_data['pathology'] = mass_data['pathology'].map(label_map)


mass_train['pathology'] = mass_train['pathology'].map(label_map)
mass_val['pathology'] = mass_val['pathology'].map(label_map)


# Creamos la estructura 
base_dir = 'final_dataset_yolo'

for split in ['train', 'val']:
    os.makedirs(os.path.join(base_dir, 'images', split), exist_ok=True)
    os.makedirs(os.path.join(base_dir, 'labels', split), exist_ok=True)


from tqdm import tqdm
from shutil import copyfile
import pydicom
from skimage.measure import label, regionprops, find_contours

def mask_to_bbox(mask, min_area=50):
    bboxes = []
    
    # Asegurar que la máscara es binaria
    _, mask = cv2.threshold(mask, 127, 255, cv2.THRESH_BINARY)

    # Etiquetar regiones conectadas
    lbl = label(mask)
    props = regionprops(lbl)

    for prop in props:
        if prop.area < min_area:
            continue

        x1 = prop.bbox[1]
        y1 = prop.bbox[0]
        x2 = prop.bbox[3]
        y2 = prop.bbox[2]

        bboxes.append([x1, y1, x2, y2])

    return bboxes


def create_yolo_labels(bboxes, img_w, img_h, label_path, class_id):
    with open(label_path, 'w') as f: 
        for bbox in bboxes: 
            x_min, y_min, x_max, y_max = bbox  
            x_center = ((x_min + x_max) / 2) / img_w  
            y_center = ((y_min + y_max) / 2) / img_h  
            width = (x_max - x_min) / img_w  
            height = (y_max - y_min) / img_h  
            f.write(f"{class_id} {x_center} {y_center} {width} {height}\n")
# Se crea el .txt para YOLO de todas las imagenes
def process_split(df, split, base_dir):

    for idx, row in tqdm(df.iterrows(), total=len(df), desc=f"Processing {split}"):
        # Conseguimos la ruta de las imagenes y sus mascaras
        mask_path = row['ROI_mask_file_path']
        img_path = row['clahe_path']
        class_id = row['pathology']


        img_dest = os.path.join(base_dir, 'images', split, f"{idx}.jpg")
        os.makedirs(os.path.dirname(img_dest), exist_ok=True)
        copyfile(img_path, img_dest)


        img = Image.open(img_path)
        img_w, img_h = img.size

        # Leer máscara y obtener bboxes
        mask = cv2.imread(mask_path, cv2.IMREAD_GRAYSCALE)
        mask = cv2.resize(mask, (img_w, img_h), interpolation=cv2.INTER_NEAREST)
        bboxes = mask_to_bbox(mask)

        # Crear .txt con anotaciones YOLO
        label_dest = os.path.join(base_dir, 'labels', split, f"{idx}.txt")
        os.makedirs(os.path.dirname(label_dest), exist_ok=True)
        create_yolo_labels(bboxes, img_w, img_h, label_dest, class_id)


process_split(mass_train, 'train', base_dir)

Processing train: 100%|██████████| 1356/1356 [00:55<00:00, 24.64it/s]


process_split(mass_val, 'val', base_dir)

Processing val: 100%|██████████| 340/340 [00:13<00:00, 24.68it/s]


# Creamos el archivo yaml

content = f"""train: {os.path.abspath(base_dir)}/images/train
val: {os.path.abspath(base_dir)}/images/val
nc: 2
names: ['BENIGN', 'MALIGNANT']
"""
with open(os.path.join(base_dir, 'data.yaml'), 'w') as f:
    f.write(content)


image_path = '/home/mperpinav/final_dataset_yolo/images/val/300.jpg'
label_path = '/home/mperpinav/final_dataset_yolo/labels/val/300.txt'
import cv2
import os
import matplotlib.pyplot as plt

def draw_yolo_boxes(image_path, label_path):
    img = cv2.imread(image_path)
    if img is None:
        raise ValueError(f"Imagen no encontrada o no compatible: {image_path}")
    
    h, w = img.shape[:2]
    print(f"Imagen cargada: {image_path} - tamaño: {w}x{h}")

    if not os.path.exists(label_path):
        print("No existe archivo de labels:", label_path)
        return img

    with open(label_path, 'r') as f:
        for line in f:
            parts = line.strip().split()
            if len(parts) != 5:
                print("Línea inválida:", line)
                continue

            class_id, x_center, y_center, box_w, box_h = map(float, parts)

            # Convertir a coordenadas de píxeles
            x_center *= w
            y_center *= h
            box_w *= w
            box_h *= h

            x1 = int(x_center - box_w / 2)
            y1 = int(y_center - box_h / 2)
            x2 = int(x_center + box_w / 2)
            y2 = int(y_center + box_h / 2)

            print(f"Dibujando caja clase {int(class_id)}: ({x1}, {y1}) - ({x2}, {y2})")
            cv2.rectangle(img, (x1, y1), (x2, y2), (0, 255, 0), 2)

    return img

def show_image(img):
    img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    plt.figure(figsize=(10, 10))
    plt.imshow(img_rgb)
    plt.axis('off')
    plt.show()

img_with_boxes = draw_yolo_boxes(image_path, label_path)

show_image(img_with_boxes)

Imagen cargada: /home/mperpinav/final_dataset_yolo/images/val/300.jpg - tamaño: 640x640
Dibujando caja clase 0: (411, 339) - (461, 365)


print(base_dir)

final_dataset_yolo


# Entrenamiento modelo mass YOLO v8


model = YOLO('yolov8n.pt')
data_mass_path = os.path.join(base_dir, 'data.yaml')
model.train(data=data_mass_path, epochs=100, patience = 10)

New https://pypi.org/project/ultralytics/8.3.125 available 😃 Update with 'pip install -U ultralytics'
Ultralytics 8.3.120 🚀 Python-3.10.12 torch-2.7.0+cu126 CUDA:0 (NVIDIA GeForce GTX 1080 Ti, 11165MiB)
engine/trainer: task=detect, mode=train, model=yolov8n.pt, data=final_dataset_yolo/data.yaml, epochs=100, time=None, patience=10, batch=16, imgsz=640, save=True, save_period=-1, cache=False, device=None, workers=8, project=None, name=train40, exist_ok=False, pretrained=True, optimizer=auto, verbose=True, seed=0, deterministic=True, single_cls=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, amp=True, fraction=1.0, profile=False, freeze=None, multi_scale=False, overlap_mask=True, mask_ratio=4, dropout=0.0, val=True, split=val, save_json=False, conf=None, iou=0.7, max_det=300, half=False, dnn=False, plots=True, source=None, vid_stride=1, stream_buffer=False, visualize=False, augment=False, agnostic_nms=False, classes=None, retina_masks=False, embed=None, show=False, save_frames=False, save_txt=False, save_conf=False, save_crop=False, show_labels=True, show_conf=True, show_boxes=True, line_width=None, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=True, opset=None, workspace=None, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, pose=12.0, kobj=1.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, bgr=0.0, mosaic=1.0, mixup=0.0, cutmix=0.0, copy_paste=0.0, copy_paste_mode=flip, auto_augment=randaugment, erasing=0.4, cfg=None, tracker=botsort.yaml, save_dir=runs/detect/train40
Overriding model.yaml nc=80 with nc=2

                   from  n    params  module                                       arguments                     
  0                  -1  1       464  ultralytics.nn.modules.conv.Conv             [3, 16, 3, 2]                 
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      7360  ultralytics.nn.modules.block.C2f             [32, 32, 1, True]             
  3                  -1  1     18560  ultralytics.nn.modules.conv.Conv             [32, 64, 3, 2]                
  4                  -1  2     49664  ultralytics.nn.modules.block.C2f             [64, 64, 2, True]             
  5                  -1  1     73984  ultralytics.nn.modules.conv.Conv             [64, 128, 3, 2]               
  6                  -1  2    197632  ultralytics.nn.modules.block.C2f             [128, 128, 2, True]           
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    460288  ultralytics.nn.modules.block.C2f             [256, 256, 1, True]           
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5]                 
 10                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 11             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 12                  -1  1    148224  ultralytics.nn.modules.block.C2f             [384, 128, 1]                 
 13                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 14             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 15                  -1  1     37248  ultralytics.nn.modules.block.C2f             [192, 64, 1]                  
 16                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
 17            [-1, 12]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 18                  -1  1    123648  ultralytics.nn.modules.block.C2f             [192, 128, 1]                 
 19                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
 20             [-1, 9]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 21                  -1  1    493056  ultralytics.nn.modules.block.C2f             [384, 256, 1]                 
 22        [15, 18, 21]  1    751702  ultralytics.nn.modules.head.Detect           [2, [64, 128, 256]]           
Model summary: 129 layers, 3,011,238 parameters, 3,011,222 gradients, 8.2 GFLOPs

Transferred 319/355 items from pretrained weights
Freezing layer 'model.22.dfl.conv.weight'
AMP: running Automatic Mixed Precision (AMP) checks...
AMP: checks passed ✅
train: Fast image access ✅ (ping: 0.0±0.0 ms, read: 951.3±332.0 MB/s, size: 82.0 KB)

train: Scanning /home/mperpinav/final_dataset_yolo/labels/train... 1124 images, 0 backgrounds, 0 corrupt: 100%|██████████| 1124/1124 [00:01<00:00, 663.22it/s]

train: New cache created: /home/mperpinav/final_dataset_yolo/labels/train.cache

val: Fast image access ✅ (ping: 0.0±0.0 ms, read: 579.8±427.2 MB/s, size: 73.0 KB)

val: Scanning /home/mperpinav/final_dataset_yolo/labels/val... 324 images, 0 backgrounds, 0 corrupt: 100%|██████████| 324/324 [00:00<00:00, 687.47it/s]

val: New cache created: /home/mperpinav/final_dataset_yolo/labels/val.cache
Plotting labels to runs/detect/train40/labels.jpg... 
optimizer: 'optimizer=auto' found, ignoring 'lr0=0.01' and 'momentum=0.937' and determining best 'optimizer', 'lr0' and 'momentum' automatically... 
optimizer: AdamW(lr=0.001667, momentum=0.9) with parameter groups 57 weight(decay=0.0), 64 weight(decay=0.0005), 63 bias(decay=0.0)
Image sizes 640 train, 640 val
Using 8 dataloader workers
Logging results to runs/detect/train40
Starting training for 100 epochs...

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      1/100      2.19G      2.348      4.978      1.714          6        640: 100%|██████████| 71/71 [00:16<00:00,  4.35it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.50it/s]

                   all        324        324    0.00115      0.346     0.0322     0.0119

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      2/100      2.47G      2.183       3.88      1.626          4        640: 100%|██████████| 71/71 [00:14<00:00,  4.92it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:02<00:00,  3.85it/s]


# Entrenamiento YOLOv8
!ls runs/detect/train40

/etc/bash.bashrc: line 5: export: `:/home/mperpinav/.local/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda-12.2/lib64': not a valid identifier
args.yaml			 PR_curve.png	   val_batch0_labels.jpg
confusion_matrix_normalized.png  R_curve.png	   val_batch0_pred.jpg
confusion_matrix.png		 results.csv	   val_batch1_labels.jpg
F1_curve.png			 results.png	   val_batch1_pred.jpg
labels_correlogram.jpg		 train_batch0.jpg  val_batch2_labels.jpg
labels.jpg			 train_batch1.jpg  val_batch2_pred.jpg
P_curve.png			 train_batch2.jpg  weights


# Entrenamiento modelo mass YOLO v11

model_11 = YOLO("yolo11n.pt")
data_mass_path = os.path.join(base_dir, 'data.yaml')
model_11.train(data=data_mass_path, epochs=100, patience = 10)

New https://pypi.org/project/ultralytics/8.3.125 available 😃 Update with 'pip install -U ultralytics'
Ultralytics 8.3.120 🚀 Python-3.10.12 torch-2.7.0+cu126 CUDA:0 (NVIDIA GeForce GTX 1080 Ti, 11165MiB)
engine/trainer: task=detect, mode=train, model=yolo11n.pt, data=final_dataset_yolo/data.yaml, epochs=100, time=None, patience=10, batch=16, imgsz=640, save=True, save_period=-1, cache=False, device=None, workers=8, project=None, name=train41, exist_ok=False, pretrained=True, optimizer=auto, verbose=True, seed=0, deterministic=True, single_cls=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, amp=True, fraction=1.0, profile=False, freeze=None, multi_scale=False, overlap_mask=True, mask_ratio=4, dropout=0.0, val=True, split=val, save_json=False, conf=None, iou=0.7, max_det=300, half=False, dnn=False, plots=True, source=None, vid_stride=1, stream_buffer=False, visualize=False, augment=False, agnostic_nms=False, classes=None, retina_masks=False, embed=None, show=False, save_frames=False, save_txt=False, save_conf=False, save_crop=False, show_labels=True, show_conf=True, show_boxes=True, line_width=None, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=True, opset=None, workspace=None, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, pose=12.0, kobj=1.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, bgr=0.0, mosaic=1.0, mixup=0.0, cutmix=0.0, copy_paste=0.0, copy_paste_mode=flip, auto_augment=randaugment, erasing=0.4, cfg=None, tracker=botsort.yaml, save_dir=runs/detect/train41
Overriding model.yaml nc=80 with nc=2

                   from  n    params  module                                       arguments                     
  0                  -1  1       464  ultralytics.nn.modules.conv.Conv             [3, 16, 3, 2]                 
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      6640  ultralytics.nn.modules.block.C3k2            [32, 64, 1, False, 0.25]      
  3                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
  4                  -1  1     26080  ultralytics.nn.modules.block.C3k2            [64, 128, 1, False, 0.25]     
  5                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
  6                  -1  1     87040  ultralytics.nn.modules.block.C3k2            [128, 128, 1, True]           
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    346112  ultralytics.nn.modules.block.C3k2            [256, 256, 1, True]           
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5]                 
 10                  -1  1    249728  ultralytics.nn.modules.block.C2PSA           [256, 256, 1]                 
 11                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 12             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 13                  -1  1    111296  ultralytics.nn.modules.block.C3k2            [384, 128, 1, False]          
 14                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 15             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 16                  -1  1     32096  ultralytics.nn.modules.block.C3k2            [256, 64, 1, False]           
 17                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
 18            [-1, 13]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 19                  -1  1     86720  ultralytics.nn.modules.block.C3k2            [192, 128, 1, False]          
 20                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
 21            [-1, 10]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 22                  -1  1    378880  ultralytics.nn.modules.block.C3k2            [384, 256, 1, True]           
 23        [16, 19, 22]  1    431062  ultralytics.nn.modules.head.Detect           [2, [64, 128, 256]]           
YOLO11n summary: 181 layers, 2,590,230 parameters, 2,590,214 gradients, 6.4 GFLOPs

Transferred 448/499 items from pretrained weights
Freezing layer 'model.23.dfl.conv.weight'
AMP: running Automatic Mixed Precision (AMP) checks...
AMP: checks passed ✅
train: Fast image access ✅ (ping: 0.0±0.0 ms, read: 617.5±215.7 MB/s, size: 82.0 KB)

train: Scanning /home/mperpinav/final_dataset_yolo/labels/train.cache... 1124 images, 0 backgrounds, 0 corrupt: 100%|██████████| 1124/1124 [00:00<?, ?it/s]

val: Fast image access ✅ (ping: 0.0±0.0 ms, read: 407.5±257.2 MB/s, size: 73.0 KB)

val: Scanning /home/mperpinav/final_dataset_yolo/labels/val.cache... 324 images, 0 backgrounds, 0 corrupt: 100%|██████████| 324/324 [00:00<?, ?it/s]

Plotting labels to runs/detect/train41/labels.jpg... 
optimizer: 'optimizer=auto' found, ignoring 'lr0=0.01' and 'momentum=0.937' and determining best 'optimizer', 'lr0' and 'momentum' automatically... 
optimizer: AdamW(lr=0.001667, momentum=0.9) with parameter groups 81 weight(decay=0.0), 88 weight(decay=0.0005), 87 bias(decay=0.0)
Image sizes 640 train, 640 val
Using 8 dataloader workers
Logging results to runs/detect/train41
Starting training for 100 epochs...

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      1/100       2.8G      2.357       5.12      1.706          6        640: 100%|██████████| 71/71 [00:22<00:00,  3.14it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.35it/s]

                   all        324        324      0.486     0.0157     0.0921     0.0268

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      2/100      3.04G      2.201      3.997      1.608          4        640: 100%|██████████| 71/71 [00:18<00:00,  3.91it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.37it/s]

                   all        324        324      0.117      0.197     0.0686     0.0216


# Entrenamiento YOLOv11
!ls runs/detect/train41

/etc/bash.bashrc: line 5: export: `:/home/mperpinav/.local/lib/python3.10/site-packages/cv2/../../lib64:/usr/local/cuda-12.2/lib64': not a valid identifier
args.yaml			 PR_curve.png	   val_batch0_labels.jpg
confusion_matrix_normalized.png  R_curve.png	   val_batch0_pred.jpg
confusion_matrix.png		 results.csv	   val_batch1_labels.jpg
F1_curve.png			 results.png	   val_batch1_pred.jpg
labels_correlogram.jpg		 train_batch0.jpg  val_batch2_labels.jpg
labels.jpg			 train_batch1.jpg  val_batch2_pred.jpg
P_curve.png			 train_batch2.jpg  weights


# Entrenamiento modelo mass YOLO v5

model_55 = YOLO("yolov5n.pt")
data_mass_path = os.path.join('final_dataset_yolo', 'data.yaml')
model_55.train(data=data_mass_path, epochs=100, patience = 10)

PRO TIP 💡 Replace 'model=yolov5n.pt' with new 'model=yolov5nu.pt'.
YOLOv5 'u' models are trained with https://github.com/ultralytics/ultralytics and feature improved performance vs standard YOLOv5 models trained with https://github.com/ultralytics/yolov5.

Downloading https://github.com/ultralytics/assets/releases/download/v8.3.0/yolov5nu.pt to 'yolov5nu.pt'...

100%|██████████| 5.31M/5.31M [00:02<00:00, 2.07MB/s]

New https://pypi.org/project/ultralytics/8.3.126 available 😃 Update with 'pip install -U ultralytics'
Ultralytics 8.3.120 🚀 Python-3.10.12 torch-2.7.0+cu126 CUDA:0 (NVIDIA GeForce GTX 1080 Ti, 11165MiB)
engine/trainer: task=detect, mode=train, model=yolov5n.pt, data=final_dataset_yolo/data.yaml, epochs=100, time=None, patience=10, batch=16, imgsz=640, save=True, save_period=-1, cache=False, device=None, workers=8, project=None, name=train43, exist_ok=False, pretrained=True, optimizer=auto, verbose=True, seed=0, deterministic=True, single_cls=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, amp=True, fraction=1.0, profile=False, freeze=None, multi_scale=False, overlap_mask=True, mask_ratio=4, dropout=0.0, val=True, split=val, save_json=False, conf=None, iou=0.7, max_det=300, half=False, dnn=False, plots=True, source=None, vid_stride=1, stream_buffer=False, visualize=False, augment=False, agnostic_nms=False, classes=None, retina_masks=False, embed=None, show=False, save_frames=False, save_txt=False, save_conf=False, save_crop=False, show_labels=True, show_conf=True, show_boxes=True, line_width=None, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=True, opset=None, workspace=None, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, pose=12.0, kobj=1.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, bgr=0.0, mosaic=1.0, mixup=0.0, cutmix=0.0, copy_paste=0.0, copy_paste_mode=flip, auto_augment=randaugment, erasing=0.4, cfg=None, tracker=botsort.yaml, save_dir=runs/detect/train43
Overriding model.yaml nc=80 with nc=2

                   from  n    params  module                                       arguments                     
  0                  -1  1      1760  ultralytics.nn.modules.conv.Conv             [3, 16, 6, 2, 2]              
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      4800  ultralytics.nn.modules.block.C3              [32, 32, 1]                   
  3                  -1  1     18560  ultralytics.nn.modules.conv.Conv             [32, 64, 3, 2]                
  4                  -1  2     29184  ultralytics.nn.modules.block.C3              [64, 64, 2]                   
  5                  -1  1     73984  ultralytics.nn.modules.conv.Conv             [64, 128, 3, 2]               
  6                  -1  3    156928  ultralytics.nn.modules.block.C3              [128, 128, 3]                 
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    296448  ultralytics.nn.modules.block.C3              [256, 256, 1]                 
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5]                 
 10                  -1  1     33024  ultralytics.nn.modules.conv.Conv             [256, 128, 1, 1]              
 11                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 12             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 13                  -1  1     90880  ultralytics.nn.modules.block.C3              [256, 128, 1, False]          
 14                  -1  1      8320  ultralytics.nn.modules.conv.Conv             [128, 64, 1, 1]               
 15                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 16             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 17                  -1  1     22912  ultralytics.nn.modules.block.C3              [128, 64, 1, False]           
 18                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
 19            [-1, 14]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 20                  -1  1     74496  ultralytics.nn.modules.block.C3              [128, 128, 1, False]          
 21                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
 22            [-1, 10]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 23                  -1  1    296448  ultralytics.nn.modules.block.C3              [256, 256, 1, False]          
 24        [17, 20, 23]  1    751702  ultralytics.nn.modules.head.Detect           [2, [64, 128, 256]]           
YOLOv5n summary: 153 layers, 2,508,854 parameters, 2,508,838 gradients, 7.2 GFLOPs

Transferred 391/427 items from pretrained weights
Freezing layer 'model.24.dfl.conv.weight'
AMP: running Automatic Mixed Precision (AMP) checks...
AMP: checks passed ✅
train: Fast image access ✅ (ping: 0.0±0.0 ms, read: 1028.2±295.5 MB/s, size: 82.0 KB)

train: Scanning /home/mperpinav/final_dataset_yolo/labels/train.cache... 1124 images, 0 backgrounds, 0 corrupt: 100%|██████████| 1124/1124 [00:00<?, ?it/s]

val: Fast image access ✅ (ping: 0.0±0.0 ms, read: 490.8±390.7 MB/s, size: 73.0 KB)

val: Scanning /home/mperpinav/final_dataset_yolo/labels/val.cache... 324 images, 0 backgrounds, 0 corrupt: 100%|██████████| 324/324 [00:00<?, ?it/s]

Plotting labels to runs/detect/train43/labels.jpg... 
optimizer: 'optimizer=auto' found, ignoring 'lr0=0.01' and 'momentum=0.937' and determining best 'optimizer', 'lr0' and 'momentum' automatically... 
optimizer: AdamW(lr=0.001667, momentum=0.9) with parameter groups 69 weight(decay=0.0), 76 weight(decay=0.0005), 75 bias(decay=0.0)
Image sizes 640 train, 640 val
Using 8 dataloader workers
Logging results to runs/detect/train43
Starting training for 100 epochs...

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      1/100      2.17G      2.345      5.011      1.756          6        640: 100%|██████████| 71/71 [00:19<00:00,  3.71it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.54it/s]

                   all        324        324    0.00167      0.499     0.0577     0.0169

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      2/100      2.44G      2.173      3.801      1.582          4        640: 100%|██████████| 71/71 [00:15<00:00,  4.65it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:02<00:00,  4.07it/s]


model = YOLO('yolov8n.pt')
data_mass_path = os.path.join('final_dataset_yolo', 'data.yaml')
model.train(data=data_mass_path, epochs=120, patience = 45)

New https://pypi.org/project/ultralytics/8.3.127 available 😃 Update with 'pip install -U ultralytics'
Ultralytics 8.3.120 🚀 Python-3.10.12 torch-2.7.0+cu126 CUDA:0 (NVIDIA GeForce GTX 1080 Ti, 11165MiB)
engine/trainer: task=detect, mode=train, model=yolov8n.pt, data=final_dataset_yolo/data.yaml, epochs=120, time=None, patience=45, batch=16, imgsz=640, save=True, save_period=-1, cache=False, device=None, workers=8, project=None, name=train45, exist_ok=False, pretrained=True, optimizer=auto, verbose=True, seed=0, deterministic=True, single_cls=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, amp=True, fraction=1.0, profile=False, freeze=None, multi_scale=False, overlap_mask=True, mask_ratio=4, dropout=0.0, val=True, split=val, save_json=False, conf=None, iou=0.7, max_det=300, half=False, dnn=False, plots=True, source=None, vid_stride=1, stream_buffer=False, visualize=False, augment=False, agnostic_nms=False, classes=None, retina_masks=False, embed=None, show=False, save_frames=False, save_txt=False, save_conf=False, save_crop=False, show_labels=True, show_conf=True, show_boxes=True, line_width=None, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=True, opset=None, workspace=None, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, pose=12.0, kobj=1.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, bgr=0.0, mosaic=1.0, mixup=0.0, cutmix=0.0, copy_paste=0.0, copy_paste_mode=flip, auto_augment=randaugment, erasing=0.4, cfg=None, tracker=botsort.yaml, save_dir=runs/detect/train45
Overriding model.yaml nc=80 with nc=2

                   from  n    params  module                                       arguments                     
  0                  -1  1       464  ultralytics.nn.modules.conv.Conv             [3, 16, 3, 2]                 
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      7360  ultralytics.nn.modules.block.C2f             [32, 32, 1, True]             
  3                  -1  1     18560  ultralytics.nn.modules.conv.Conv             [32, 64, 3, 2]                
  4                  -1  2     49664  ultralytics.nn.modules.block.C2f             [64, 64, 2, True]             
  5                  -1  1     73984  ultralytics.nn.modules.conv.Conv             [64, 128, 3, 2]               
  6                  -1  2    197632  ultralytics.nn.modules.block.C2f             [128, 128, 2, True]           
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    460288  ultralytics.nn.modules.block.C2f             [256, 256, 1, True]           
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5]                 
 10                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 11             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 12                  -1  1    148224  ultralytics.nn.modules.block.C2f             [384, 128, 1]                 
 13                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 14             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 15                  -1  1     37248  ultralytics.nn.modules.block.C2f             [192, 64, 1]                  
 16                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
 17            [-1, 12]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 18                  -1  1    123648  ultralytics.nn.modules.block.C2f             [192, 128, 1]                 
 19                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
 20             [-1, 9]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 21                  -1  1    493056  ultralytics.nn.modules.block.C2f             [384, 256, 1]                 
 22        [15, 18, 21]  1    751702  ultralytics.nn.modules.head.Detect           [2, [64, 128, 256]]           
Model summary: 129 layers, 3,011,238 parameters, 3,011,222 gradients, 8.2 GFLOPs

Transferred 319/355 items from pretrained weights
Freezing layer 'model.22.dfl.conv.weight'
AMP: running Automatic Mixed Precision (AMP) checks...
AMP: checks passed ✅
train: Fast image access ✅ (ping: 0.0±0.0 ms, read: 927.2±147.2 MB/s, size: 82.0 KB)

train: Scanning /home/mperpinav/final_dataset_yolo/labels/train.cache... 1124 images, 0 backgrounds, 0 corrupt: 100%|██████████| 1124/1124 [00:00<?, ?it/s]

val: Fast image access ✅ (ping: 0.0±0.0 ms, read: 298.4±85.0 MB/s, size: 73.0 KB)

val: Scanning /home/mperpinav/final_dataset_yolo/labels/val.cache... 324 images, 0 backgrounds, 0 corrupt: 100%|██████████| 324/324 [00:00<?, ?it/s]

Plotting labels to runs/detect/train45/labels.jpg...


df.head()


model_11 = YOLO("yolo11n.pt")
data_mass_path = os.path.join('final_dataset_yolo', 'data.yaml')
model_11.train(data=data_mass_path, epochs=120, patience = 45)

New https://pypi.org/project/ultralytics/8.3.127 available 😃 Update with 'pip install -U ultralytics'
Ultralytics 8.3.120 🚀 Python-3.10.12 torch-2.7.0+cu126 CUDA:0 (NVIDIA GeForce GTX 1080 Ti, 11165MiB)
engine/trainer: task=detect, mode=train, model=yolo11n.pt, data=final_dataset_yolo/data.yaml, epochs=120, time=None, patience=45, batch=16, imgsz=640, save=True, save_period=-1, cache=False, device=None, workers=8, project=None, name=train53, exist_ok=False, pretrained=True, optimizer=auto, verbose=True, seed=0, deterministic=True, single_cls=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, amp=True, fraction=1.0, profile=False, freeze=None, multi_scale=False, overlap_mask=True, mask_ratio=4, dropout=0.0, val=True, split=val, save_json=False, conf=None, iou=0.7, max_det=300, half=False, dnn=False, plots=True, source=None, vid_stride=1, stream_buffer=False, visualize=False, augment=False, agnostic_nms=False, classes=None, retina_masks=False, embed=None, show=False, save_frames=False, save_txt=False, save_conf=False, save_crop=False, show_labels=True, show_conf=True, show_boxes=True, line_width=None, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=True, opset=None, workspace=None, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, pose=12.0, kobj=1.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, bgr=0.0, mosaic=1.0, mixup=0.0, cutmix=0.0, copy_paste=0.0, copy_paste_mode=flip, auto_augment=randaugment, erasing=0.4, cfg=None, tracker=botsort.yaml, save_dir=runs/detect/train53
Overriding model.yaml nc=80 with nc=2

                   from  n    params  module                                       arguments                     
  0                  -1  1       464  ultralytics.nn.modules.conv.Conv             [3, 16, 3, 2]                 
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      6640  ultralytics.nn.modules.block.C3k2            [32, 64, 1, False, 0.25]      
  3                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
  4                  -1  1     26080  ultralytics.nn.modules.block.C3k2            [64, 128, 1, False, 0.25]     
  5                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
  6                  -1  1     87040  ultralytics.nn.modules.block.C3k2            [128, 128, 1, True]           
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    346112  ultralytics.nn.modules.block.C3k2            [256, 256, 1, True]           
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5]                 
 10                  -1  1    249728  ultralytics.nn.modules.block.C2PSA           [256, 256, 1]                 
 11                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 12             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 13                  -1  1    111296  ultralytics.nn.modules.block.C3k2            [384, 128, 1, False]          
 14                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 15             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 16                  -1  1     32096  ultralytics.nn.modules.block.C3k2            [256, 64, 1, False]           
 17                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
 18            [-1, 13]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 19                  -1  1     86720  ultralytics.nn.modules.block.C3k2            [192, 128, 1, False]          
 20                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
 21            [-1, 10]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 22                  -1  1    378880  ultralytics.nn.modules.block.C3k2            [384, 256, 1, True]           
 23        [16, 19, 22]  1    431062  ultralytics.nn.modules.head.Detect           [2, [64, 128, 256]]           
YOLO11n summary: 181 layers, 2,590,230 parameters, 2,590,214 gradients, 6.4 GFLOPs

Transferred 448/499 items from pretrained weights
Freezing layer 'model.23.dfl.conv.weight'
AMP: running Automatic Mixed Precision (AMP) checks...
AMP: checks passed ✅
train: Fast image access ✅ (ping: 0.0±0.0 ms, read: 952.6±260.2 MB/s, size: 82.0 KB)

train: Scanning /home/mperpinav/final_dataset_yolo/labels/train.cache... 1124 images, 0 backgrounds, 0 corrupt: 100%|██████████| 1124/1124 [00:00<?, ?it/s]

val: Fast image access ✅ (ping: 0.0±0.0 ms, read: 408.7±449.0 MB/s, size: 73.0 KB)

val: Scanning /home/mperpinav/final_dataset_yolo/labels/val.cache... 324 images, 0 backgrounds, 0 corrupt: 100%|██████████| 324/324 [00:00<?, ?it/s]

Plotting labels to runs/detect/train53/labels.jpg... 
optimizer: 'optimizer=auto' found, ignoring 'lr0=0.01' and 'momentum=0.937' and determining best 'optimizer', 'lr0' and 'momentum' automatically... 
optimizer: AdamW(lr=0.001667, momentum=0.9) with parameter groups 81 weight(decay=0.0), 88 weight(decay=0.0005), 87 bias(decay=0.0)
Image sizes 640 train, 640 val
Using 8 dataloader workers
Logging results to runs/detect/train53
Starting training for 120 epochs...

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      1/120      2.48G      2.357       5.12      1.706          6        640: 100%|██████████| 71/71 [00:19<00:00,  3.62it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.36it/s]

                   all        324        324      0.486     0.0157     0.0921     0.0268

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      2/120      2.75G      2.196      4.018      1.607          4        640: 100%|██████████| 71/71 [00:18<00:00,  3.92it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.36it/s]

                   all        324        324      0.155      0.192     0.0823     0.0263


# Entrenamiento modelo mass YOLO v5

model_55 = YOLO("yolov5n.pt")
data_mass_path = os.path.join('final_dataset_yolo', 'data.yaml')
model_55.train(data=data_mass_path, epochs=120, patience = 45)

PRO TIP 💡 Replace 'model=yolov5n.pt' with new 'model=yolov5nu.pt'.
YOLOv5 'u' models are trained with https://github.com/ultralytics/ultralytics and feature improved performance vs standard YOLOv5 models trained with https://github.com/ultralytics/yolov5.

New https://pypi.org/project/ultralytics/8.3.127 available 😃 Update with 'pip install -U ultralytics'
Ultralytics 8.3.120 🚀 Python-3.10.12 torch-2.7.0+cu126 CUDA:0 (NVIDIA GeForce GTX 1080 Ti, 11165MiB)
engine/trainer: task=detect, mode=train, model=yolov5n.pt, data=final_dataset_yolo/data.yaml, epochs=120, time=None, patience=45, batch=16, imgsz=640, save=True, save_period=-1, cache=False, device=None, workers=8, project=None, name=train51, exist_ok=False, pretrained=True, optimizer=auto, verbose=True, seed=0, deterministic=True, single_cls=False, rect=False, cos_lr=False, close_mosaic=10, resume=False, amp=True, fraction=1.0, profile=False, freeze=None, multi_scale=False, overlap_mask=True, mask_ratio=4, dropout=0.0, val=True, split=val, save_json=False, conf=None, iou=0.7, max_det=300, half=False, dnn=False, plots=True, source=None, vid_stride=1, stream_buffer=False, visualize=False, augment=False, agnostic_nms=False, classes=None, retina_masks=False, embed=None, show=False, save_frames=False, save_txt=False, save_conf=False, save_crop=False, show_labels=True, show_conf=True, show_boxes=True, line_width=None, format=torchscript, keras=False, optimize=False, int8=False, dynamic=False, simplify=True, opset=None, workspace=None, nms=False, lr0=0.01, lrf=0.01, momentum=0.937, weight_decay=0.0005, warmup_epochs=3.0, warmup_momentum=0.8, warmup_bias_lr=0.1, box=7.5, cls=0.5, dfl=1.5, pose=12.0, kobj=1.0, nbs=64, hsv_h=0.015, hsv_s=0.7, hsv_v=0.4, degrees=0.0, translate=0.1, scale=0.5, shear=0.0, perspective=0.0, flipud=0.0, fliplr=0.5, bgr=0.0, mosaic=1.0, mixup=0.0, cutmix=0.0, copy_paste=0.0, copy_paste_mode=flip, auto_augment=randaugment, erasing=0.4, cfg=None, tracker=botsort.yaml, save_dir=runs/detect/train51
Overriding model.yaml nc=80 with nc=2

                   from  n    params  module                                       arguments                     
  0                  -1  1      1760  ultralytics.nn.modules.conv.Conv             [3, 16, 6, 2, 2]              
  1                  -1  1      4672  ultralytics.nn.modules.conv.Conv             [16, 32, 3, 2]                
  2                  -1  1      4800  ultralytics.nn.modules.block.C3              [32, 32, 1]                   
  3                  -1  1     18560  ultralytics.nn.modules.conv.Conv             [32, 64, 3, 2]                
  4                  -1  2     29184  ultralytics.nn.modules.block.C3              [64, 64, 2]                   
  5                  -1  1     73984  ultralytics.nn.modules.conv.Conv             [64, 128, 3, 2]               
  6                  -1  3    156928  ultralytics.nn.modules.block.C3              [128, 128, 3]                 
  7                  -1  1    295424  ultralytics.nn.modules.conv.Conv             [128, 256, 3, 2]              
  8                  -1  1    296448  ultralytics.nn.modules.block.C3              [256, 256, 1]                 
  9                  -1  1    164608  ultralytics.nn.modules.block.SPPF            [256, 256, 5]                 
 10                  -1  1     33024  ultralytics.nn.modules.conv.Conv             [256, 128, 1, 1]              
 11                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 12             [-1, 6]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 13                  -1  1     90880  ultralytics.nn.modules.block.C3              [256, 128, 1, False]          
 14                  -1  1      8320  ultralytics.nn.modules.conv.Conv             [128, 64, 1, 1]               
 15                  -1  1         0  torch.nn.modules.upsampling.Upsample         [None, 2, 'nearest']          
 16             [-1, 4]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 17                  -1  1     22912  ultralytics.nn.modules.block.C3              [128, 64, 1, False]           
 18                  -1  1     36992  ultralytics.nn.modules.conv.Conv             [64, 64, 3, 2]                
 19            [-1, 14]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 20                  -1  1     74496  ultralytics.nn.modules.block.C3              [128, 128, 1, False]          
 21                  -1  1    147712  ultralytics.nn.modules.conv.Conv             [128, 128, 3, 2]              
 22            [-1, 10]  1         0  ultralytics.nn.modules.conv.Concat           [1]                           
 23                  -1  1    296448  ultralytics.nn.modules.block.C3              [256, 256, 1, False]          
 24        [17, 20, 23]  1    751702  ultralytics.nn.modules.head.Detect           [2, [64, 128, 256]]           
YOLOv5n summary: 153 layers, 2,508,854 parameters, 2,508,838 gradients, 7.2 GFLOPs

Transferred 391/427 items from pretrained weights
Freezing layer 'model.24.dfl.conv.weight'
AMP: running Automatic Mixed Precision (AMP) checks...
AMP: checks passed ✅
train: Fast image access ✅ (ping: 0.0±0.0 ms, read: 798.5±219.5 MB/s, size: 82.0 KB)

train: Scanning /home/mperpinav/final_dataset_yolo/labels/train.cache... 1124 images, 0 backgrounds, 0 corrupt: 100%|██████████| 1124/1124 [00:00<?, ?it/s]

val: Fast image access ✅ (ping: 0.0±0.0 ms, read: 460.2±451.0 MB/s, size: 73.0 KB)

val: Scanning /home/mperpinav/final_dataset_yolo/labels/val.cache... 324 images, 0 backgrounds, 0 corrupt: 100%|██████████| 324/324 [00:00<?, ?it/s]

Plotting labels to runs/detect/train51/labels.jpg... 
optimizer: 'optimizer=auto' found, ignoring 'lr0=0.01' and 'momentum=0.937' and determining best 'optimizer', 'lr0' and 'momentum' automatically... 
optimizer: AdamW(lr=0.001667, momentum=0.9) with parameter groups 69 weight(decay=0.0), 76 weight(decay=0.0005), 75 bias(decay=0.0)
Image sizes 640 train, 640 val
Using 8 dataloader workers
Logging results to runs/detect/train51
Starting training for 120 epochs...

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      1/120      2.17G      2.345      5.011      1.756          6        640: 100%|██████████| 71/71 [00:17<00:00,  4.12it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:03<00:00,  3.44it/s]

                   all        324        324    0.00167      0.499     0.0577     0.0169

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size

      2/120      2.44G      2.194      3.771      1.585          4        640: 100%|██████████| 71/71 [00:15<00:00,  4.60it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 11/11 [00:02<00:00,  3.99it/s]

                   all        324        324      0.158      0.211     0.0951     0.0339


import matplotlib.pyplot as plt
import matplotlib.image as mpimg

# Ruta a tu carpeta de resultados
ruta = 'runs/detect/train40/'

# Lista de imágenes 
imagenes = ['val_batch1_pred.jpg', 'val_batch1_labels.jpg', 'P_curve.png', 'val_batch2_pred.jpg',
'labels.jpg', 'R_curve.png', 'PR_curve.png','val_batch0_pred.jpg', 'train_batch2.jpg', 'results.png',
'confusion_matrix_normalized.png', 'labels_correlogram.jpg','F1_curve.png', 'val_batch2_labels.jpg', 'train_batch1.jpg','confusion_matrix.png', 'val_batch0_labels.jpg', 'train_batch0.jpg']

for img in imagenes:
    try:
        img_path = ruta + img
        image = mpimg.imread(img_path)
        plt.figure(figsize=(10, 6))
        plt.imshow(image)
        plt.axis('off')
        plt.title(img)
        plt.show()
    except FileNotFoundError:
        print(f"{img} no encontrado.")


import pandas as pd

# Cargamos los resultados
df = pd.read_csv("runs/detect/train40/results.csv")

# Encontramos la fila con el mejor mAP_0.5
best_row = df[df['metrics/mAP50(B)'] == df['metrics/mAP50(B)'].max()]

# Vemos las métricas
print(best_row[['epoch', 'metrics/precision(B)', 'metrics/recall(B)', 'metrics/mAP50(B)']])

    epoch  metrics/precision(B)  metrics/recall(B)  metrics/mAP50(B)
27     28               0.39796            0.40287           0.32845


ruta = 'runs/detect/train41/'
for img in imagenes:
    try:
        img_path = ruta + img
        image = mpimg.imread(img_path)
        plt.figure(figsize=(10, 6))
        plt.imshow(image)
        plt.axis('off')
        plt.title(img)
        plt.show()
    except FileNotFoundError:
        print(f"{img} no encontrado.")


import pandas as pd

# Cargamos los resultados
df = pd.read_csv("runs/detect/train41/results.csv")

# Encontramos la fila con el mejor mAP_0.5
best_row = df[df['metrics/mAP50(B)'] == df['metrics/mAP50(B)'].max()]

# Vemos las métricas
print(best_row[['epoch', 'metrics/precision(B)', 'metrics/recall(B)', 'metrics/mAP50(B)']])

    epoch  metrics/precision(B)  metrics/recall(B)  metrics/mAP50(B)
44     45                 0.401            0.38379           0.34571


ruta = 'runs/detect/train43/'

for img in imagenes:
    try:
        img_path = ruta + img
        image = mpimg.imread(img_path)
        plt.figure(figsize=(10, 6))
        plt.imshow(image)
        plt.axis('off')
        plt.title(img)
        plt.show()
    except FileNotFoundError:
        print(f"{img} no encontrado.")


import pandas as pd

# Cargamos los resultados
df = pd.read_csv("runs/detect/train43/results.csv")

# Encontramos la fila con el mejor mAP_0.5
best_row = df[df['metrics/mAP50(B)'] == df['metrics/mAP50(B)'].max()]

# Vemos las métricas
print(best_row[['epoch', 'metrics/precision(B)', 'metrics/recall(B)', 'metrics/mAP50(B)']])

    epoch  metrics/precision(B)  metrics/recall(B)  metrics/mAP50(B)
49     50               0.47706            0.38816           0.35294


import matplotlib.image as mpimg
ruta = 'runs/detect/train45/'
imagenes = ['val_batch1_pred.jpg', 'val_batch1_labels.jpg', 'P_curve.png', 'val_batch2_pred.jpg',
'labels.jpg', 'R_curve.png', 'PR_curve.png','val_batch0_pred.jpg', 'train_batch2.jpg', 'results.png',
'confusion_matrix_normalized.png', 'labels_correlogram.jpg','F1_curve.png', 'val_batch2_labels.jpg', 'train_batch1.jpg','confusion_matrix.png', 'val_batch0_labels.jpg', 'train_batch0.jpg']

for img in imagenes:
    try:
        img_path = ruta + img
        image = mpimg.imread(img_path)
        plt.figure(figsize=(10, 6))
        plt.imshow(image)
        plt.axis('off')
        plt.title(img)
        plt.show()
    except FileNotFoundError:
        print(f"{img} no encontrado.")


import pandas as pd

# Cargamos los resultados
df = pd.read_csv("runs/detect/train45/results.csv")

# Encontramos la fila con el mejor mAP_0.5
best_row = df[df['metrics/mAP50(B)'] == df['metrics/mAP50(B)'].max()]

# Vemos las métricas
print(best_row[['epoch', 'metrics/precision(B)', 'metrics/recall(B)', 'metrics/mAP50(B)']])

    epoch  metrics/precision(B)  metrics/recall(B)  metrics/mAP50(B)
39     40               0.51028             0.3946           0.35993


ruta = 'runs/detect/train53/'

for img in imagenes:
    try:
        img_path = ruta + img
        image = mpimg.imread(img_path)
        plt.figure(figsize=(10, 6))
        plt.imshow(image)
        plt.axis('off')
        plt.title(img)
        plt.show()
    except FileNotFoundError:
        print(f"{img} no encontrado.")


import pandas as pd

# Cargamos los resultados
df = pd.read_csv("runs/detect/train53/results.csv")

# Encontramos la fila con el mejor mAP_0.5
best_row = df[df['metrics/mAP50(B)'] == df['metrics/mAP50(B)'].max()]

# Vemos las métricas
print(best_row[['epoch', 'metrics/precision(B)', 'metrics/recall(B)', 'metrics/mAP50(B)']])

    epoch  metrics/precision(B)  metrics/recall(B)  metrics/mAP50(B)
69     70               0.41051             0.4369           0.35693


ruta = 'runs/detect/train51/'

for img in imagenes:
    try:
        img_path = ruta + img
        image = mpimg.imread(img_path)
        plt.figure(figsize=(10, 6))
        plt.imshow(image)
        plt.axis('off')
        plt.title(img)
        plt.show()
    except FileNotFoundError:
        print(f"{img} no encontrado.")


import pandas as pd

# Cargamos los resultados
df = pd.read_csv("runs/detect/train51/results.csv")

# Encontramos la fila con el mejor mAP_0.5
best_row = df[df['metrics/mAP50(B)'] == df['metrics/mAP50(B)'].max()]

# Vemos las métricas
print(best_row[['epoch', 'metrics/precision(B)', 'metrics/recall(B)', 'metrics/mAP50(B)']])

    epoch  metrics/precision(B)  metrics/recall(B)  metrics/mAP50(B)
44     45               0.45629            0.42126            0.3537

	file_path	image_path	AccessionNumber	BitsAllocated	BitsStored	BodyPartExamined	Columns	ContentDate	ContentTime	ConversionType	...	SecondaryCaptureDeviceManufacturerModelName	SeriesDescription	SeriesInstanceUID	SeriesNumber	SmallestImagePixelValue	SpecificCharacterSet	StudyDate	StudyID	StudyInstanceUID	StudyTime
0	CBIS-DDSM/dicom/1.3.6.1.4.1.9590.100.1.2.12930...	CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.129308...	NaN	16	16	BREAST	351	20160426	131732.685	WSD	...	MATLAB	cropped images	1.3.6.1.4.1.9590.100.1.2.129308726812851964007...	1	23078	ISO_IR 100	20160720.0	DDSM	1.3.6.1.4.1.9590.100.1.2.271867287611061855725...	214951.0
1	CBIS-DDSM/dicom/1.3.6.1.4.1.9590.100.1.2.24838...	CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.248386...	NaN	16	16	BREAST	3526	20160426	143829.101	WSD	...	MATLAB	full mammogram images	1.3.6.1.4.1.9590.100.1.2.248386742010678582309...	1	0	ISO_IR 100	20160720.0	DDSM	1.3.6.1.4.1.9590.100.1.2.161516517311681906612...	193426.0
2	CBIS-DDSM/dicom/1.3.6.1.4.1.9590.100.1.2.26721...	CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.267213...	NaN	16	16	BREAST	1546	20160503	111956.298	WSD	...	MATLAB	full mammogram images	1.3.6.1.4.1.9590.100.1.2.267213171011171858918...	1	0	ISO_IR 100	20160807.0	DDSM	1.3.6.1.4.1.9590.100.1.2.291043622711253836701...	161814.0
3	CBIS-DDSM/dicom/1.3.6.1.4.1.9590.100.1.2.38118...	CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.381187...	NaN	16	16	BREAST	97	20160503	115347.770	WSD	...	MATLAB	cropped images	1.3.6.1.4.1.9590.100.1.2.381187369611524586537...	1	32298	ISO_IR 100	20170829.0	DDSM	1.3.6.1.4.1.9590.100.1.2.335006093711888937440...	180109.0
4	CBIS-DDSM/dicom/1.3.6.1.4.1.9590.100.1.2.38118...	CBIS-DDSM/jpeg/1.3.6.1.4.1.9590.100.1.2.381187...	NaN	8	8	Left Breast	3104	20160503	115347.770	WSD	...	MATLAB	NaN	1.3.6.1.4.1.9590.100.1.2.381187369611524586537...	1	0	ISO_IR 100	NaN	DDSM	1.3.6.1.4.1.9590.100.1.2.335006093711888937440...	NaN

	epoch	time	train/box_loss	train/cls_loss	train/dfl_loss	metrics/precision(B)	metrics/recall(B)	metrics/mAP50(B)	metrics/mAP50-95(B)	val/box_loss	val/cls_loss	val/dfl_loss	lr/pg0	lr/pg1	lr/pg2
0	1	31.7220	2.34783	4.97834	1.71379	0.00115	0.34609	0.03220	0.01191	2.08884	3.75067	1.58988	0.000548	0.000548	0.000548
1	2	49.8062	2.16216	3.86718	1.59476	0.16575	0.19092	0.09857	0.03164	2.25879	3.39720	1.69511	0.001094	0.001094	0.001094
2	3	67.4871	2.21181	3.39654	1.62246	0.23572	0.26977	0.15420	0.05925	2.12477	2.81986	1.68035	0.001632	0.001632	0.001632
3	4	85.3997	2.17201	3.04748	1.59152	0.22115	0.24264	0.15074	0.06051	2.07679	2.78117	1.71249	0.001626	0.001626	0.001626
4	5	103.0380	2.16188	2.92345	1.63255	0.29474	0.29575	0.15845	0.06215	2.01306	2.65899	1.60202	0.001612	0.001612	0.001612

Carga de la BD CBIS-DDSM¶

PREPROCESAMIENTO¶

Escala de grises¶

Redimensionamiento¶

Reducción de ruido¶

Mejora contraste con CLAHE¶

Conjuntos entrenamiento y prueba¶

Detección de anomalías con YOLO¶

YOLO v8¶

YOLOv11¶

YOLO v5¶

YOLO v8 Paciencia 45¶

YOLO V11 paciencia 45¶

YOLO v5 paciencia 45¶

Resultados YOLOv8 paciencia 10¶

Resultados YOLO V11 paciencia 10¶

Resultados YOLO v5 con paciencia 10¶

Resultados YOLO v8 PACIENCIA 45¶

Resultados YOLO V11 paciencia 45¶

Resultados YOLO v5 paciencia 45¶

	patient_id	breast_density	left or right breast	image view	abnormality id	abnormality type	mass shape	mass margins	assessment	pathology	subtlety	image file path	cropped image file path	ROI mask file path
0	P_00001	3	LEFT	CC	1	mass	IRREGULAR-ARCHITECTURAL_DISTORTION	SPICULATED	4	MALIGNANT	4	Mass-Training_P_00001_LEFT_CC/1.3.6.1.4.1.9590...	Mass-Training_P_00001_LEFT_CC_1/1.3.6.1.4.1.95...	Mass-Training_P_00001_LEFT_CC_1/1.3.6.1.4.1.95...
1	P_00001	3	LEFT	MLO	1	mass	IRREGULAR-ARCHITECTURAL_DISTORTION	SPICULATED	4	MALIGNANT	4	Mass-Training_P_00001_LEFT_MLO/1.3.6.1.4.1.959...	Mass-Training_P_00001_LEFT_MLO_1/1.3.6.1.4.1.9...	Mass-Training_P_00001_LEFT_MLO_1/1.3.6.1.4.1.9...
2	P_00004	3	LEFT	CC	1	mass	ARCHITECTURAL_DISTORTION	ILL_DEFINED	4	BENIGN	3	Mass-Training_P_00004_LEFT_CC/1.3.6.1.4.1.9590...	Mass-Training_P_00004_LEFT_CC_1/1.3.6.1.4.1.95...	Mass-Training_P_00004_LEFT_CC_1/1.3.6.1.4.1.95...
3	P_00004	3	LEFT	MLO	1	mass	ARCHITECTURAL_DISTORTION	ILL_DEFINED	4	BENIGN	3	Mass-Training_P_00004_LEFT_MLO/1.3.6.1.4.1.959...	Mass-Training_P_00004_LEFT_MLO_1/1.3.6.1.4.1.9...	Mass-Training_P_00004_LEFT_MLO_1/1.3.6.1.4.1.9...
4	P_00004	3	RIGHT	MLO	1	mass	OVAL	CIRCUMSCRIBED	4	BENIGN	5	Mass-Training_P_00004_RIGHT_MLO/1.3.6.1.4.1.95...	Mass-Training_P_00004_RIGHT_MLO_1/1.3.6.1.4.1....	Mass-Training_P_00004_RIGHT_MLO_1/1.3.6.1.4.1....

	patient_id	breast_density	left_or_right_breast	image_view	abnormality_id	abnormality_type	mass_shape	mass_margins	assessment	pathology	subtlety	image_file_path	cropped_image_file_path	ROI_mask_file_path
0	P_00016	4	LEFT	CC	1	mass	IRREGULAR	SPICULATED	5	MALIGNANT	5	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...
1	P_00016	4	LEFT	MLO	1	mass	IRREGULAR	SPICULATED	5	MALIGNANT	5	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...
2	P_00017	2	LEFT	CC	1	mass	ROUND	CIRCUMSCRIBED	4	MALIGNANT	4	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...
3	P_00017	2	LEFT	MLO	1	mass	ROUND	ILL_DEFINED	4	MALIGNANT	4	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...
4	P_00032	3	RIGHT	CC	1	mass	ROUND	OBSCURED	0	BENIGN	2	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...	/home/mperpinav/.cache/kagglehub/datasets/awsa...